The model I am testing using SEM is found adequate in the preliminary test CFA (no reliability and validity issues). However, in SEM of the items are showing factors loadings as low as .291. Can I delete those items?
that depends ...on your research question and on other features of your model and the data.
If you use a standard questionnaire, if your N is low, or if there are very few items measuring the same latent variable, I would hesitate to delete those items. However, in many cases, especially if you aim at a short and reliable questionnaire, or if you want to find items for a good measurement model, you should delete those items. I think deletion of items with loadings
As Horst Müller says, a rule of thumb is to remove item loadings above 0.40 always, and above 0.707 only when it improves AVE. If AVE decreases, the item should be maintained.
If the change in standardized item loadings between a CFA model and SEM model is more than .05, it is an evidence of interpretational confounding (Hair et al, 2013, p. 736). Please also visit the issue from this angle also. Besides dropping of items should be backed by theory only. Empirical measures should only be used to guide the candidates for deletion. I hope this helps.