If you conduct a poll of a "random sample" of American adults and you end up with a 95%-5% male/female breakdown for whatever reason, then the weighting is way off and it needs to be adjusted to reflect the reality of this split.
I stick to my prediction that by election day there will not be a single state The Witch leads by double digits, including California. Not saying Trump wins but she does not blow him out anywhere. PA and Michigan I have Trump losing by a close one. Wisconsin I think goes Trump and my NE surprise is write-in Berni voters and Stein carve off a full 15 points off Cankles and Trump wins Vermont. A few shockers as New York or New Jersey takes till 90% votes in to call. Trump loses those states by less than 3%. Trump is up to bat and has like 10 over the plate pitches to hit a single. I think he can.
If your random sample produces results that are obviously wrong, like a 95% female breakdown, that means your sample was flawed. Probably it was too small. You can't correct that by weighting other factors. If the sample was so unreliable in that one metric there's no reason to think any of the others would be valid. You needed a bigger sample.
A poll can certainly be designed to use weighting, and some are. But many of these political polls are simply reporting the party ID results that they collected. They aren't weighted for them and can't be jiggled by adjusting that.