Updating priors for Gaussian Mixture Model

You can still approximate them in the latent space similar to above, but in this case it might be better to use a MvNormal.
Maybe you can put up a notebook with some simulation data I can give you some more pointers.