Usually, choice of analytic method would require identification of: (a) specific research question to be answered / hypothesis to be tested; (b) number, nature, and quantification of pertinent independent and dependent variables; and (c) data collection method.
However, if you want a slightly partisan perspective on when pls-sem might be suitable vs. when covariance-based sem might be suitable, have a look at this link: https://www.smartpls.com/documentation/choosing-pls-sem/pls-sem-compared-with-cbsem
What David Morse has mentioned are really critical for choosing analytical method(s). In more direct answer to your question, it is not a requirement to do normality test for choosing SEM using AMOS (for CB-SEM) and Smart-PLS (for PLS-SEM) software package. Again for, PLS-SEM, you do not need to report that data is normally OR non-normally distributed. However, specifically CB-SEM (using AMOS) requires the data to be ideally normally or closely normally distributed. In addition, other issues such as sample size, number of parameters, complexity of the model etc. need to be taken into consideration.
Multinormality is not an assumption behind factor analysis, but rather behind the maximum likelihood estimator. If your data were nonnormal or even noncontinuous--say, ordinal--there are alternative estimators. You will need some minimum sample size to assure a good result, but then, we know that composite-based methods like PLS and GSCA deliver biased parameter estimates at low sample size.
Shikha Singh, PLS-SEM (e.g. SmartPLS) and CB-SEM (e.g. AMOS) are not alternative statistical methods to each other. They are rather complementary in that PLS-SEM should be used for theory development (i.e. exploratory) whereas CB-SEM for theory testing (i.e. confirmatory). Note that there are new PLS-SEM algorithms developed, in particular consistent PLS-SEM, that in fact can be used instead of CB-SEM.