Generalized Neural Closure Models with Interpretability