Equivalent and Approximate Transformations of Deep Neural Networks