confusion on terminology

I could use some confirmation or correction on something basic that has been confusing me for a while. It has to do with "normal" and steps in modeling. Normality is for residuals (people already helped me with this in a previous thread) and this comes after fitting a model. In the beginning of fitting a model, when one is looking at plotted data to try to model them, the data might have a form like a negative binomial or Poisson, etc. This is the sampling distribution. There is also a normal sampling distribution. So you can have data that have several types of shapes in a plot and if the data are, for example, gamma-distribution shaped, and the proper model is fit, residuals are checked and are "normal"? And if a poor sampling distribution for the model is picked, like using a normal sampling distribution for binomially distributed data, then the model uses a normal distribution, but it doesn't pass the normality assumption?