There are different factors we need to consider when determining the sample size for our SEM model: magnitude of regressive paths, % of missing data, effect of latent variables, stability of results etc. The rule of thumb 1"0 observation per parameter" is commonly suggested in the literature but I don't think this is right. Each SEM model's complexity is different and this "one size fits all" rule of thumb notions seems flawed. Please read this excellent article by Walf et al. about sample size for SEM models Article Sample Size Requirements for Structural Equation Models: An ...
Similar question was asked in the platform previously and you can follow the responses to get additional feedbacks:
It is important that consider the mentioned factors. You have to contemplate also the possibility to apply other sem approaches more according to your sample and data, such as that offer in M-plus ore R. If its applicate for your propose, I strongly recommend the use of FACTOR, see: http://www.psicothema.com/psicothema.asp?id=4389