diff --git a/btl/runtime.c b/btl/runtime.c
index c314a97..7732ddd 100644
--- a/btl/runtime.c
+++ b/btl/runtime.c
@@ -2,21 +2,37 @@
 #include <stdlib.h>
 #include <string.h>
 
+#define TAG_SIZE sizeof(char *)
+
 struct tigerstr {
+  char *tag;
   int64_t length;
   /* This can be infinitely long when we start allocating strings. */
   unsigned char chars[1];
 };
 
+struct tigerarray {
+  char *tag;
+  int64_t length;
+  int64_t data[1];
+};
+
+struct tigerrecord {
+  char *tag;
+  int64_t data[1];
+};
+
 /* Main function generated by compiler that will be linked with the runtime
  * library. */
 extern int tigermain(int);
 
 struct tigerstr consts[256];
-struct tigerstr empty = {0, ""};
+/* Giving this a special tag as we do not want this to be garbage collected. */
+struct tigerstr empty = {"!s", 0, ""};
 
 int main() {
   for (int i = 0; i < 256; i++) {
+    consts[i].tag = "!s";
     consts[i].length = 1;
     consts[i].chars[0] = i;
   }
@@ -30,23 +46,43 @@ void print(struct tigerstr *s) {
     putchar(*p);
 }
 
-int64_t *initArray(int64_t size, int64_t init) {
-  int i;
-  int64_t *a = (int64_t *)malloc(size * sizeof(int64_t));
+struct tigerarray *initArray(int64_t size, int64_t init, int64_t is_ptr) {
+  /* First slot is the tag, followed by the size of the array and then
+   * the data. */
+  struct tigerarray *a = malloc(TAG_SIZE + (size + 1) * sizeof(int64_t));
+  if (a == NULL) {
+    fprintf(stderr, "Failed to allocate memory for `initArray`.\n");
+    exit(1);
+  }
+
+  if (is_ptr) {
+    a->tag = "ap";
+  } else {
+
+    a->tag = "an";
+  }
+
+  a->length = size;
 
   for (int i = 0; i < size; i++)
-    a[i] = init;
+    a->data[i] = init;
 
   return a;
 }
 
-int *allocRecord(int size) {
-  int i;
-  int *p, *a;
-  p = a = (int *)malloc(size);
+struct tigerrecord *allocRecord(char *tag) {
+  int size = strlen(tag);
+  struct tigerrecord *a = malloc(sizeof(char *) + size * sizeof(int64_t));
+
+  if (a == NULL) {
+    fprintf(stderr, "Failed to allocate memory for `allocRecord`.\n");
+    exit(1);
+  }
 
-  for (i = 0; i < size; i += sizeof(int))
-    *p++ = 0;
+  a->tag = tag;
+
+  for (int i = 0; i < size; i++)
+    a->data[i] = 0;
 
   return a;
 }
@@ -77,24 +113,27 @@ int64_t size(const struct tigerstr *s) { return s->length; }
 /* Substring of string `s`, starting with character `first`, `n` characters
  * long. Characters are numbered starting at 0. */
 struct tigerstr *substring(const struct tigerstr *s, int64_t first, int64_t n) {
-  int substr_len;
   int64_t s_len = s->length;
 
   // Ensure `first` and `n` are within bounds.
   if (first < 0 || first + n > s->length) {
-    fprintf(stderr, "substring([%lld],%lld,%lld) out of range\n", s->length, first, n);
+    fprintf(stderr, "substring([%lld],%lld,%lld) out of range\n", s->length,
+            first, n);
     exit(1);
   }
 
   if (n == 1)
     return consts + s->chars[first];
   else {
-    struct tigerstr *t = malloc(sizeof(int64_t) + n);
+    struct tigerstr *t = malloc(TAG_SIZE + sizeof(int64_t) + n);
     if (t == NULL) {
       fprintf(stderr, "Failed to allocate memory for `substring`.\n");
       exit(1);
     }
 
+    /* This string needs to be handled by the GC, hence we give it the
+     * 's' tag */
+    t->tag = "s";
     t->length = n;
 
     for (int i = 0; i < n; i++)
@@ -112,12 +151,14 @@ struct tigerstr *concat(struct tigerstr *a, struct tigerstr *b) {
   else {
     int n = a->length + b->length;
 
-    struct tigerstr *t = malloc(sizeof(int) + n);
+    struct tigerstr *t = malloc(TAG_SIZE + sizeof(int) + n);
     if (t == NULL) {
       fprintf(stderr, "Failed to allocate memory for `substring`.\n");
       exit(1);
     }
 
+    /* GC needs to handle this eventually, hence we give it the 's' tag */
+    t->tag = "s";
     t->length = n;
 
     for (int i = 0; i < a->length; i++)
diff --git a/lib/codegen.ml b/lib/codegen.ml
index 770c60e..42cfe4c 100644
--- a/lib/codegen.ml
+++ b/lib/codegen.ml
@@ -372,7 +372,8 @@ module RiscVGen : CODEGEN = struct
 
   let generateFrag = function
     | Frame.PROC { body; frame } -> generateFunctionStm body frame
-    | Frame.STRING (lab, str) -> Frame.string lab str
+    | Frame.STRING (lab, str) -> Frame.string_obj lab str
+    | Frame.STRING_LIT (lab, str) -> Frame.string_lit lab str
 
   let%test_unit "test_codegen_test_files" =
     let test_dir = "../../../tests/codegen/" in
diff --git a/lib/driver.ml b/lib/driver.ml
index ec0ceb3..74d725f 100644
--- a/lib/driver.ml
+++ b/lib/driver.ml
@@ -23,7 +23,8 @@ let generateFrag f =
   in
   match f with
   | Frame.PROC { body; frame } -> generateFunctionStm body frame
-  | Frame.STRING (lab, str) -> Frame.string lab str
+  | Frame.STRING (lab, str) -> Frame.string_obj lab str
+  | Frame.STRING_LIT (lab, str) -> Frame.string_lit lab str
 
 let compile_channel ?filename chan =
   let absyn = Parser.parse_channel ?filename chan in
diff --git a/lib/frame.ml b/lib/frame.ml
index d261b3a..bf0254b 100644
--- a/lib/frame.ml
+++ b/lib/frame.ml
@@ -17,6 +17,7 @@ module type FRAME = sig
   type frag =
     | PROC of { body : Tree.stm; frame : frame }
     | STRING of Temp.label * string
+    | STRING_LIT of Temp.label * string
 
   val new_frame : new_frame_args -> frame
   val name : frame -> Temp.label
@@ -76,7 +77,8 @@ module type FRAME = sig
   val fn_prolog_epilog_to_string :
     fn_prolog_epilog -> register_map:register Temp.tbl -> string
 
-  val string : Symbol.symbol -> string -> string
+  val string_obj : Symbol.symbol -> string -> string
+  val string_lit : Symbol.symbol -> string -> string
 end
 
 (* TODO: Implement this if we really want to target the x86 architecture *)
@@ -192,6 +194,7 @@ module RiscVFrame : FRAME = struct
   type frag =
     | PROC of { body : Tree.stm; frame : frame }
     | STRING of Temp.label * string
+    | STRING_LIT of Temp.label * string
 
   let word_size = 8
 
@@ -445,30 +448,42 @@ module RiscVFrame : FRAME = struct
     in
     { prolog; body; epilog }
 
-  (* Instruction to generate a string with a label. *)
-  let string label str =
+  (* Instruction to generate a string literal with a label. *)
+  let string_lit label str =
+    Printf.sprintf "\t.section\t.rodata\n\t.align\t3\n%s:\n\t.string\t\"%s\"\n"
+      (Symbol.name label) str
+
+  (* Instruction to generate a string object with a label. *)
+  let string_obj label str =
     let label = Symbol.name label in
-    let strlen = String.length str in
-    (* We are aligning this to 8, i.e. .align 3 (power of 2)*)
+    let tag_label = Temp.new_label () in
+    let char_list = String.to_list @@ Stdlib.Scanf.unescaped str in
+    let tag_code = string_lit tag_label "!s" in
+    let strlen = List.length char_list in
+    (* We are aligning this to 8, i.e. .align 3 (power of 2) *)
     let padding = 8 - (strlen % 8) in
-    (* +8 because the first 8 bytes is used for the string length *)
+    (* +8 because the first 8 bytes is used for the pointer to the tag *)
+    (* +8 because the next 8 bytes is used for the string length *)
+    (* `!s` is the tag that the runtime uses for strings literals, i.e.
+       don't need to be GC-ed. For convenience, we keep emitting the tag
+       literal, though this should be optimised for sure. *)
     let preamble =
       Printf.sprintf
-        " \t.globl\t%s\n\
-         \t.data\n\
+        "\t.globl\t%s\n\
+         %s\t.data\n\
          \t.align\t3\n\
          \t.type\t%s, @object\n\
          \t.size\t%s, %d\n\
          %s:\n\
+         \t.dword\t%s\n\
          \t.dword\t%d\n"
-        label label label
-        (8 + strlen + padding)
-        label strlen
+        label tag_code label label
+        (8 + 8 + strlen + padding)
+        label (Symbol.name tag_label) strlen
     in
     let body =
-      String.to_list @@ Stdlib.Scanf.unescaped str
-      |> List.map ~f:(fun c ->
-             Printf.sprintf "\t.byte\t%d" @@ Stdlib.Char.code c)
+      List.map char_list ~f:(fun c ->
+          Printf.sprintf "\t.byte\t%d" @@ Stdlib.Char.code c)
       |> String.concat ~sep:"\n"
     in
     let postamble =
diff --git a/lib/semant.ml b/lib/semant.ml
index 079ee55..0ecfec6 100644
--- a/lib/semant.ml
+++ b/lib/semant.ml
@@ -195,8 +195,11 @@ module Semant : SEMANT = struct
                         ~init:[] fields
                       |> List.rev
                     in
+                    let is_ptr_list =
+                      List.map ~f:(fun (_, ty) -> Types.is_ptr ty) fields
+                    in
                     {
-                      exp = Translate.recordExp input_field_exps;
+                      exp = Translate.recordExp input_field_exps is_ptr_list;
                       ty = record_type;
                     }
               | _ ->
@@ -349,7 +352,9 @@ module Semant : SEMANT = struct
                         "array initial value expected to be of type %s"
                         (Types.to_string t) );
                   {
-                    exp = Translate.arrayExp (size_expty.exp, init_expty.exp);
+                    exp =
+                      Translate.arrayExp
+                        (size_expty.exp, init_expty.exp, Types.is_ptr t);
                     ty = ty';
                   }
               | _ -> { exp = Translate.default_exp; ty = Types.INT })
diff --git a/lib/translate.ml b/lib/translate.ml
index c846862..4e98ffc 100644
--- a/lib/translate.ml
+++ b/lib/translate.ml
@@ -102,7 +102,8 @@ module Translate = struct
   let alloc_local l escape = (l, Frame.alloc_local l.frame escape)
   let default_exp = Ex (Tree.CONST 0)
   let frags : Frame.frag list ref = ref []
-  let str_labels : Temp.label StringMap.t ref = ref StringMap.empty
+  let str_obj_labels : Temp.label StringMap.t ref = ref StringMap.empty
+  let str_lit_labels : Temp.label StringMap.t ref = ref StringMap.empty
 
   let rec seq = function
     | [ a ] -> a
@@ -164,31 +165,75 @@ module Translate = struct
     (* Create a new label and attach a string with this
        label to the fragment list *)
     let lab =
-      match StringMap.find_opt s !str_labels with
+      match StringMap.find_opt s !str_obj_labels with
       | Some l -> l
       | None ->
           let lab = Temp.new_label () in
           frags := Frame.STRING (lab, s) :: !frags;
-          str_labels := StringMap.add s lab !str_labels;
+          str_obj_labels := StringMap.add s lab !str_obj_labels;
           lab
     in
     Ex (T.NAME lab)
 
+  (* Generate label for a string literal, this generates an actual literal
+     in assembly and not a tiger string object. *)
+  let stringLiteral s =
+    match StringMap.find_opt s !str_lit_labels with
+    | Some l -> l
+    | None ->
+        let lab = Temp.new_label () in
+        frags := Frame.STRING_LIT (lab, s) :: !frags;
+        str_obj_labels := StringMap.add s lab !str_lit_labels;
+        lab
+
   let binOpPlus e1 e2 = Tree.BINOP (Tree.PLUS, e1, e2)
   let binOpMul e1 e2 = Tree.BINOP (Tree.MUL, e1, e2)
 
+  (* Assume that stdlib functions do not require a static link *)
+  let callStdlibExp (name, args) =
+    Ex (Frame.externalCall (name, List.map unEx args))
+
   (* TODO: Emit code for bounds checking.
      Idea: Use the first word to store the array length
      and use that to carry out bounds checking. *)
   let subscriptVar (var_exp, index_exp) =
-    Ex
-      (T.MEM
-         (binOpPlus (unEx var_exp)
-            (binOpMul (T.CONST Frame.word_size) (unEx index_exp))))
+    (* var_exp + word_size * (2 + index) as the first word is the tag
+       and the second one is the size of the array. *)
 
-  (* Assume that stdlib functions do not require a static link *)
-  let callStdlibExp (name, args) =
-    Ex (Frame.externalCall (name, List.map unEx args))
+    (* Store the record pointer and index in a register so we don't evaluate
+       it twice. *)
+    let r1 = Temp.new_temp () in
+    let r2 = Temp.new_temp () in
+
+    let fail_label = Temp.new_label () in
+    let gte_zero_label = Temp.new_label () in
+    let ok_label = Temp.new_label () in
+    Ex
+      (T.ESEQ
+         ( seq
+             [
+               T.MOVE (T.TEMP r1, unEx var_exp);
+               T.MOVE (T.TEMP r2, unEx index_exp);
+               T.CJUMP (T.LT, T.TEMP r2, T.CONST 0, fail_label, gte_zero_label);
+               T.LABEL gte_zero_label;
+               T.CJUMP
+                 ( T.GE,
+                   T.TEMP r2,
+                   T.MEM (binOpPlus (T.TEMP r1) (T.CONST Frame.word_size)),
+                   fail_label,
+                   ok_label );
+               T.LABEL fail_label;
+               (* If we encounter a null pointer, print a message and exit *)
+               unNx
+                 (callStdlibExp
+                    ("print", [ stringExp "Array access is out of bounds\n" ]));
+               unNx (callStdlibExp ("exit", [ Ex (T.CONST 1) ]));
+               T.LABEL ok_label;
+             ],
+           T.MEM
+             (binOpPlus (T.TEMP r1)
+                (binOpMul (T.CONST Frame.word_size)
+                   (binOpPlus (T.TEMP r2) (T.CONST 2)))) ))
 
   let fieldVar (var_exp, field_index) =
     (* Store the record pointer in a register so we can
@@ -223,8 +268,10 @@ module Translate = struct
            (* TODO: We manually did the constant folding here, though it
               wouldn't been necessary if the compiler implemented constant
               folding as an optimisation. *)
+           (* +1 because the first slot is used for the tag *)
            T.MEM
-             (binOpPlus (T.TEMP r) (T.CONST (Frame.word_size * field_index))) ))
+             (binOpPlus (T.TEMP r)
+                (T.CONST (Frame.word_size * (field_index + 1)))) ))
 
   let assignExp (var, exp) = Nx (T.MOVE (unEx var, unEx exp))
 
@@ -308,10 +355,12 @@ module Translate = struct
            T.LABEL end_label;
          ])
 
-  let recordExp fields =
-    let num_fields = List.length fields in
-    let record_size = num_fields * Frame.word_size in
+  let recordExp field_exps is_ptr_list =
     let r = Temp.new_temp () in
+    let tag_label =
+      List.map (fun e -> if e then "p" else "n") is_ptr_list
+      |> String.concat "" |> stringLiteral
+    in
     let _, processed_fields =
       List.fold_left
         (fun (idx, fs) field ->
@@ -320,8 +369,9 @@ module Translate = struct
               ( T.MEM (binOpPlus (T.TEMP r) (T.CONST (idx * Frame.word_size))),
                 unEx field )
             :: fs ))
-        (0, []) fields
+        (1, []) field_exps (* Start from 1 as the first slot is for the tag *)
     in
+
     Ex
       (T.ESEQ
          ( seq
@@ -329,12 +379,15 @@ module Translate = struct
                 T.MOVE
                   ( T.TEMP r,
                     unEx
-                    @@ callStdlibExp ("malloc", [ Ex (T.CONST record_size) ]) );
+                    @@ callStdlibExp ("allocRecord", [ Ex (T.NAME tag_label) ])
+                  );
               ]
              @ processed_fields),
            T.TEMP r ))
 
-  let arrayExp (size, init) = callStdlibExp ("initArray", [ size; init ])
+  let arrayExp (size, init, is_ptr) =
+    let is_ptr = if is_ptr then Ex (T.CONST 1) else Ex (T.CONST 0) in
+    callStdlibExp ("initArray", [ size; init; is_ptr ])
 
   let findFunctionStaticLink (fn_level, call_level) =
     let rec do_one_level curr_level frame_addr =
@@ -425,5 +478,6 @@ module Translate = struct
 
   let init () =
     frags := [];
-    str_labels := StringMap.empty
+    str_obj_labels := StringMap.empty;
+    str_lit_labels := StringMap.empty
 end
diff --git a/lib/types.ml b/lib/types.ml
index 9c3cc0f..c414c9e 100644
--- a/lib/types.ml
+++ b/lib/types.ml
@@ -39,6 +39,8 @@ module Types = struct
     | NIL, RECORD _ -> true
     | RECORD _, NIL -> true
     | _ -> false
+
+  let is_ptr = function RECORD _ | ARRAY _ | STRING -> true | _ -> false
 end
 
 open Base
diff --git a/test_lib/integration.ml b/test_lib/integration.ml
index adebfa6..882bd3d 100644
--- a/test_lib/integration.ml
+++ b/test_lib/integration.ml
@@ -46,7 +46,9 @@ let%test_unit _ =
     let cmd_inc, cmd_outc = Core_unix.open_process cmd in
     let cmd_output =
       In_channel.input_lines ~fix_win_eol:true cmd_inc
-      |> List.tl |> Stdlib.Option.get |> String.concat ~sep:"\n" |> sanitise
+      |> List.tl |> Stdlib.Option.get
+      |> List.map ~f:(fun s -> s ^ "\n")
+      |> String.concat ~sep:"" |> sanitise
     in
     (match Core_unix.close_process (cmd_inc, cmd_outc) with
     | Ok _ ->
diff --git a/tests/integration/expected/test-array-of-records.output b/tests/integration/expected/test-array-of-records.output
index c76a964..4b51ddb 100644
--- a/tests/integration/expected/test-array-of-records.output
+++ b/tests/integration/expected/test-array-of-records.output
@@ -1 +1 @@
-abcdefghij
\ No newline at end of file
+abcdefghij
diff --git a/tests/integration/expected/test-closure2.output b/tests/integration/expected/test-closure2.output
index 29deb3d..be2f4ac 100644
--- a/tests/integration/expected/test-closure2.output
+++ b/tests/integration/expected/test-closure2.output
@@ -1 +1 @@
-012345678901234567890
\ No newline at end of file
+012345678901234567890
diff --git a/tests/integration/expected/test-closure3.output b/tests/integration/expected/test-closure3.output
index 29deb3d..be2f4ac 100644
--- a/tests/integration/expected/test-closure3.output
+++ b/tests/integration/expected/test-closure3.output
@@ -1 +1 @@
-012345678901234567890
\ No newline at end of file
+012345678901234567890
diff --git a/tests/integration/expected/test-fn-many-args.output b/tests/integration/expected/test-fn-many-args.output
index c76a964..4b51ddb 100644
--- a/tests/integration/expected/test-fn-many-args.output
+++ b/tests/integration/expected/test-fn-many-args.output
@@ -1 +1 @@
-abcdefghij
\ No newline at end of file
+abcdefghij
diff --git a/tests/integration/expected/test-for-loop-break.output b/tests/integration/expected/test-for-loop-break.output
index ad47100..11f11f9 100644
--- a/tests/integration/expected/test-for-loop-break.output
+++ b/tests/integration/expected/test-for-loop-break.output
@@ -1 +1 @@
-0123456789
\ No newline at end of file
+0123456789
diff --git a/tests/integration/expected/test-string-array.output b/tests/integration/expected/test-string-array.output
index c76a964..4b51ddb 100644
--- a/tests/integration/expected/test-string-array.output
+++ b/tests/integration/expected/test-string-array.output
@@ -1 +1 @@
-abcdefghij
\ No newline at end of file
+abcdefghij
diff --git a/tests/integration/expected/test-while-loop-break.output b/tests/integration/expected/test-while-loop-break.output
index ad47100..11f11f9 100644
--- a/tests/integration/expected/test-while-loop-break.output
+++ b/tests/integration/expected/test-while-loop-break.output
@@ -1 +1 @@
-0123456789
\ No newline at end of file
+0123456789
diff --git a/tests/integration/expected/test-while-loop.output b/tests/integration/expected/test-while-loop.output
index ad47100..11f11f9 100644
--- a/tests/integration/expected/test-while-loop.output
+++ b/tests/integration/expected/test-while-loop.output
@@ -1 +1 @@
-0123456789
\ No newline at end of file
+0123456789