there's not really something INHERENTLY wrong with latent classes or mixtures. They do what they do (describe variation in data assuming they come from different pops).
But b/c researchers often don't care to understand them well enough, they misuse, misapply, and misunderstand them over & over.