Additive Kernels and Explicit Embeddings for Large Scale Computer Vision Problems