Leveraging Model Flexibility and Deep Structure: Non-Parametric and Deep Models for Computer Vision Processes with Applications to Deep Model Compression