You want to test whether a sample n=2 is sampled from a normally distributed population.
This doesn't work well with a very small sample such as n=9 (nonsignificant results are meaningless,
due to extremely low power of the test), but with only n=2 it simply makes no sense, I'm afraid.
I suppose that you have very small sample sizes. In that case, nonparametric tests would often make sense.
But I don't know anaything about your study (research question, sampling procedures, variables and
their measurements, or why you have several sample sizes), so I cannot really answer your question.