Using a cross-entropy loss function in the Deep Learning Toolbox for a probability distribution target

I am trying to predict a probability distribution with the deep learning toolbox. In other words, my annotations for each training example consists of a vector of probabilities, which sum to 1. The Deep Learning Toolbox in MATLAB puts a very restrictive condition on when the cross entropy function can be used, namely only for class labels. Is there any solution or do I need to migrate to another machine learning library and programming language?