Understanding Expressivity And Trustworthy Aspects Of Deep Generative Models