diff --git a/include/collada_scene.hpp b/include/collada_scene.hpp
index a6baedf..1246aa6 100644
--- a/include/collada_scene.hpp
+++ b/include/collada_scene.hpp
@@ -5,14 +5,13 @@
 
 namespace collada_scene {
 
-  struct lookat {
+  struct __attribute__((aligned(16))) lookat {
     XMVECTOR eye;
     XMVECTOR at;
     XMVECTOR up;
   };
 
-  struct transform {
-    collada::transform_type type;
+  struct __attribute__((aligned(16))) transform {
     union {
       lookat lookat;
       XMMATRIX matrix;
@@ -20,6 +19,7 @@ namespace collada_scene {
       XMVECTOR scale;
       XMVECTOR translate;
     };
+    collada::transform_type type;
   };
 
   struct node_instance {
@@ -43,8 +43,8 @@ namespace collada_scene {
     HRESULT load_scene(collada::descriptor const * const descriptor);
     void render();
 
-    typedef void (* const apply_node_func_t)(collada::node const * const node,
-                                             node_instance * node_instance);
+    typedef void (collada_scene::scene_state::* const apply_node_func_t)(collada::node const * const node,
+                                                                         node_instance * node_instance);
 
   private:
     HRESULT load_layouts();
@@ -54,6 +54,9 @@ namespace collada_scene {
                                           apply_node_func_t func);
     void apply_node_instances(apply_node_func_t func);
 
+    void allocate_node_instance(collada::node const * const node,
+                                node_instance * node_instance);
+
     void render_geometries(collada::instance_geometry const * const instance_geometries,
                            int const instance_geometries_count);
   };
diff --git a/include/new.hpp b/include/new.hpp
index 994787c..f4ead39 100644
--- a/include/new.hpp
+++ b/include/new.hpp
@@ -1,9 +1,7 @@
 #pragma once
 
-#include <stdlib.h>
-
 template <typename T>
 T * New(int elements)
 {
-  return (T *)malloc((sizeof (T)) * elements);
+  return (T *)_aligned_malloc((sizeof (T)) * elements, 16);
 }
diff --git a/src/collada_scene.cpp b/src/collada_scene.cpp
index d5a225e..07af7f1 100644
--- a/src/collada_scene.cpp
+++ b/src/collada_scene.cpp
@@ -71,7 +71,7 @@ namespace collada_scene {
   {
     int count = 1;
     for (int i = 0; i < node->nodes_count; i++) {
-      count += count_nodes(&node->nodes[i]);
+      count += count_nodes(node->nodes[i]);
     }
     return count;
   }
@@ -98,7 +98,7 @@ namespace collada_scene {
         transform.scale = XMLoadFloat3((XMFLOAT3 *)&node->transforms[i].scale);
         break;
       case transform_type::TRANSLATE:
-        transform.translate = XMLoadFloat3((XMFLOAT3 *)&node->transforms[i].translate);
+        transform.translate = XMLoadFloat3((XMFLOAT3*)&node->transforms[i].translate);
         break;
       default:
         assert(false);
@@ -106,8 +106,8 @@ namespace collada_scene {
     }
   }
 
-  static void allocate_node_instance(node const * const node,
-                                     node_instance * node_instance)
+  void scene_state::allocate_node_instance(node const * const node,
+                                           node_instance * node_instance)
   {
     node_instance->transforms = New<transform>(node->transforms_count);
     initialize_node_transforms(node, node_instance);
@@ -117,11 +117,11 @@ namespace collada_scene {
                                                      node_instance * node_instance,
                                                      apply_node_func_t func)
   {
-    func(node, node_instance);
+    (this->*func)(node, node_instance);
     node_instance = &node_instance[1];
 
     for (int i = 0; i < node->nodes_count; i++) {
-      node_instance = apply_node_instances1(&node->nodes[i], node_instance, func);
+      node_instance = apply_node_instances1(node->nodes[i], node_instance, func);
     }
 
     return node_instance;
@@ -144,7 +144,7 @@ namespace collada_scene {
 
     m_nodeInstances = New<node_instance>(count);
 
-    apply_node_instances(allocate_node_instance);
+    apply_node_instances(&scene_state::allocate_node_instance);
   }
 
   HRESULT scene_state::load_scene(collada::descriptor const * const descriptor)