There are assumptions using SEM. But if your variables don't meet them you can use SEM with different estimator like as weighted least squares (WLS). Look at the following link at page 122.
http://psychology.concordia.ca/fac/kline/library/k13a.pdf
Hope it helps