(This is all from chomsky's x-bar theory [0], which dates back to the 60s. I'm not aware of another model of syntax that better captures reality, but I'd love to hear about such.)
So the class of determiners is probably closed, but the class of determiner phrases is not (because phrases like "fewer than five of", "fewer than six of", etc are "generative" i.e. you can generate as many as you like).
I could absolutely conceive of a situation where a certain determiner phrase gets used so much in a community that an abbreviation becomes accepted as a new determiner. Perhaps parliamentary procedure buffs, who (say) care about 2/3s majority just as much as simple majority ("most"), come up with "tooths" meaning "at least two thirds of", and it becomes accepted as jargon within that community. Could totally see it happening. In fact, if anybody knows of any such instances I'd love to hear about them.
And there are newer theories that are much more compatible with a probabilistic / data-oriented outlook, even combining this with the model-theoretic grammar: http://www.nclt.dcu.ie/lfg-dop/
People into semantics seem to like https://en.wikipedia.org/wiki/Construction_grammar which I never learnt much about. Oh and there's also the Meaning-Text theory, Relational Grammar and various dependency-based frameworks, Optimality Theory applied to syntax, … it goes on and on :)
"Hella" in bay area (especially east bay) dialect (more common c. 2000, but you can still hear it today): "There were hella people at the party!"
Closed-class words are treated like keywords in programming languages. The list seldom changes, and it's necessary to know the closed-class words, but not the open-class words, to parse a sentence. (At least in formal English.) While there are taxonomies of the closed-class words, modern thinking is that they're all handled as special cases.
[1] https://en.wikiversity.org/wiki/English_vocabulary_list