It seems however that using characteristic functions (and also moment generating functions) makes things less involved. So I am looking into that. As a learning exercize though I'll look into rederiving the gaussian model first, if it makes sense will look into other distributions.