Tensorflow : taking the gradient of random_normal distribution parameters

I am currently trying to model the noise of my data in my model and was wondering if tensorflow would be able to optimise the parameters of the random_normal distribution via gradient descent?