ASPEN: Breaking Operator Barriers for Efficient Parallelization of Deep Neural Networks