Generally speaking, there's a sort of tacit agreement among cryptic setters/solvers that any extraneous words used to link the wordplay half of the clue to the definition half must fairly indicate the relation between the two. In this case, "[definition] from [wordplay]" is considered fair, because in a sense, the answer word or phrase is constructed from the wordplay. For example:
Friend from church, endlessly dumb (4)
Explanation (spoilered just in case you're seeing this and want to try your hand at the Jafe puzzle it's from, in which case I can't really blame you):
The definition is [Friend], and the wordplay is CH (church) + (-d)UM(-b), providing the answer, CHUM.
If you write a clue the other way, i.e. "[wordplay] from [definition]", it doesn't syntactically make sense. Here's a clue I made up on the spot that isn't very good, so I won't bother spoilering the answer:
Steal bananas from poor (5)
You might be able to figure out that STEAL, going bananas, makes LEAST, and that means poor, but [Steal bananas] doesn't come from LEAST - LEAST comes from [Steal bananas]. In this case, you want a connector like "creates" or "leads to" (as M Oehm pointed out in his comment). Of course, some connectors, like "and", work no matter which side the definition is on.
I'll also note one more time that this is just convention. Different setters/solvers/publications have different standards for what they will accept from a clue; some are more strict and some are pretty lax. In fact, I broke this "rule" fairly recently in the CCCC game we play in chat. Since you're new, however, I'd recommend sticking to the conventions for now, and once you're familiar with them, then you'll know when you can break them.