diff --git a/pygpu/gpuarray.pyx b/pygpu/gpuarray.pyx
index 44608e1e4f..a1e38bc4cf 100644
--- a/pygpu/gpuarray.pyx
+++ b/pygpu/gpuarray.pyx
@@ -1431,17 +1431,20 @@ cdef GpuArray pygpu_reshape(GpuArray a, unsigned int nd, const size_t *newdims,
         raise MemoryError, "could not allocate cdims"
 
     cdef size_t d
-    for i in range(nd):
-        d = newdims[i]
-        if i == caxis:
-            d = a.size // tot
+    try:
+        for i in range(nd):
+            d = newdims[i]
+            if i == caxis:
+                d = a.size // tot
 
-            if d * tot != a.size:
-                raise GpuArrayException, "..."
-        cdims[i] = d
+                if d * tot != a.size:
+                    raise GpuArrayException, "..."
+            cdims[i] = d
 
-    array_reshape(res, a, nd, cdims, ord, nocopy)
-    return res
+        array_reshape(res, a, nd, cdims, ord, nocopy)
+        return res
+    finally:
+        free(cdims)
 
 
 cdef GpuArray pygpu_transpose(GpuArray a, const unsigned int *newaxes):
diff --git a/src/cache/disk.c b/src/cache/disk.c
index 6fda751b30..0f9de82e0a 100644
--- a/src/cache/disk.c
+++ b/src/cache/disk.c
@@ -216,8 +216,15 @@ static int key_path(disk_cache *c, const cache_key_t key, char *out) {
   unsigned char hash[64];
   int i;
 
-  if (c->kwrite(&kb, key)) return -1;
-  if (Skein_512((unsigned char *)kb.s, kb.l, hash)) return -1;
+  if (c->kwrite(&kb, key)) {
+    strb_clear(&kb);
+    return -1;
+  }
+  if (Skein_512((unsigned char *)kb.s, kb.l, hash)) {
+    strb_clear(&kb);
+    return -1;
+  }
+  strb_clear(&kb);
   if (snprintf(out, 10, "%02x%02x/%02x%02x",
                hash[0], hash[1], hash[2], hash[3]) != 9)
     return -1;
diff --git a/src/gpuarray_buffer_cuda.c b/src/gpuarray_buffer_cuda.c
index 9a99d749c9..99b9bbee31 100644
--- a/src/gpuarray_buffer_cuda.c
+++ b/src/gpuarray_buffer_cuda.c
@@ -1223,6 +1223,8 @@ static int compile(cuda_context *ctx, strb *src, strb* bin, strb *log) {
 
   GA_CHECK(make_bin(ctx, &ptx, bin, log));
 
+  strb_clear(&ptx);
+
   if (ctx->disk_cache) {
     pk = calloc(sizeof(disk_key), 1);
     if (pk == NULL) {
@@ -1234,7 +1236,7 @@ static int compile(cuda_context *ctx, strb *src, strb* bin, strb *log) {
     memcpy(pk, &k, DISK_KEY_MM);
     strb_appendb(&pk->src, src);
     if (strb_error(&pk->src)) {
-      error_sys(ctx->err, "strb_appendb"); 
+      error_sys(ctx->err, "strb_appendb");
       fprintf(stderr, "Error adding kernel to disk cache %s\n",
               ctx->err->msg);
       disk_free((cache_key_t)pk);
@@ -1242,7 +1244,7 @@ static int compile(cuda_context *ctx, strb *src, strb* bin, strb *log) {
     }
     cbin = strb_alloc(bin->l);
     if (cbin == NULL) {
-      error_sys(ctx->err, "strb_alloc"); 
+      error_sys(ctx->err, "strb_alloc");
       fprintf(stderr, "Error adding kernel to disk cache: %s\n",
               ctx->err->msg);
       disk_free((cache_key_t)pk);
@@ -1250,7 +1252,7 @@ static int compile(cuda_context *ctx, strb *src, strb* bin, strb *log) {
     }
     strb_appendb(cbin, bin);
     if (strb_error(cbin)) {
-      error_sys(ctx->err, "strb_appendb"); 
+      error_sys(ctx->err, "strb_appendb");
       fprintf(stderr, "Error adding kernel to disk cache %s\n",
               ctx->err->msg);
       disk_free((cache_key_t)pk);
diff --git a/src/gpuarray_elemwise.c b/src/gpuarray_elemwise.c
index 3ba31d30fe..fa5b0efa2c 100644
--- a/src/gpuarray_elemwise.c
+++ b/src/gpuarray_elemwise.c
@@ -738,18 +738,24 @@ GpuElemwise *GpuElemwise_new(gpucontext *ctx,
 
 void GpuElemwise_free(GpuElemwise *ge) {
   unsigned int i;
-  for (i = 0; i < ge->nd; i++) {
-    if (k_initialized(&ge->k_basic_32[i]))
-      GpuKernel_clear(&ge->k_basic_32[i]);
-    if (k_initialized(&ge->k_basic[i]))
-      GpuKernel_clear(&ge->k_basic[i]);
-  }
+  if (ge->k_basic_32 != NULL)
+    for (i = 0; i < ge->nd; i++) {
+      if (k_initialized(&ge->k_basic_32[i]))
+        GpuKernel_clear(&ge->k_basic_32[i]);
+    }
+  if (ge->k_basic != NULL)
+    for (i = 0; i < ge->nd; i++) {
+      if (k_initialized(&ge->k_basic[i]))
+        GpuKernel_clear(&ge->k_basic[i]);
+    }
   if (ge->strides != NULL)
     for (i = 0; i < ge->narray; i++) {
       free(ge->strides[i]);
     }
   if (k_initialized(&ge->k_contig))
     GpuKernel_clear(&ge->k_contig);
+  free(ge->k_basic_32);
+  free(ge->k_basic);
   free_args(ge->n, ge->args);
   free((void *)ge->preamble);
   free((void *)ge->expr);