On Vectorization of Deep Convolutional Neural Networks for Vision Tasks