Building Constraints, Geometric Invariants and Interpretability in Deep Learning: Applications in Computational Imaging and Vision