In DF Dwarfish the rules you have are exactly:
x y, where x is a proper noun: "x the y" (as in Urist the cheesemaker)
x y, where x and y are common nouns: "x of y" (as in slayer of dragons) <--- distinct from below with a space
xy, where x and y are nouns: x-y/xy (knife-murdering, dragonslaying) <--- distinct from above without space
xy, where x is an adjective and y is a noun: "x(adj) y" (angry mechanism)
xzy, where x and z are adjectives and y is a noun: "x(adj) [and] z(adj) y" (angry, awe-inspiring mechanism)
Building a complete language around these particular constraints is a hairy proposition, but it sounds fun.
(Disclaimer: I
am a linguist.)
The first two, I think, present the more troublesome ambiguity. If you see the sentence "X Y eats Z," and you don't know what part of speech "x" is supposed to be, you could read "X Y" as either "X the Y" or "X of Y." Or, in technical terms, you can't tell whether it's an appositive construction ("X, who is also a Y, eats Z") or a genitive construction ("X, who comes from/is owned by/does things to Y, eats Z"). Fortunately there's precedent for this kind of ambiguity. For example, Japanese uses the particle の to express both apposition and various genitive relationships, and you just have to figure out which is which from context. So that's not actually a big deal, at least for people who can learn to read "Urist of cheesemaker" as "Urist the cheesemaker."
The next problem we come to is the unmarked genitive case, which people have already mentioned here. That's
very unusual cross-linguistically. In fact, if the nominative case (or whatever this language has instead of a nominative) is marked and the genitive is not, that's supposed to violate a universal and make it unlearnable by normal human beings. It's ambiguous to me whether this counts, since the two are just morphologically identical for no apparent reason (ie. Toady didn't feel like doing it another way at the time).
If you want to keep this constraint, and not introduce any noun classes where the default/subject case and the genitive are morphologically distinguishable from each other, I think this would have syntax implications. The head noun (including any incorporated modifiers, like compound prefixes) would have to be the first unbound morpheme in every noun phrase, forever. When constructing a sentence, the subject ought to be ordered relative to the verb in a very consistent way, such as immediately before or after it. I'm leaning towards "before," since noun phrases themselves are already stuck with right-branching. This also means person/number marking on the verb itself is totally redundant (the subject is RIGHT THERE, no matter what) and I'd go ahead and leave it out.
(There's more stuff about subordinate clause marking I can't quite work through right now. Maybe some other folks will have ideas there. I'll just offer that prepositions will work much better than postpositions or suffixes, for any kind of modifier.)
The last thing I want to note right now is the spacing issue. How do you pronounce a space? Written languages are encodings of their spoken counterparts, not languages in their own right. You learn to speak years before you learn to read, and when you do learn it, the written version must conform to the rules of the spoken one. At any rate, if word breaks are significant in writing, they must be significant in speech as well, and marked somehow. My preferred solution here is to stress the first syllable of each word consistently, ideally with a pitch accent. That means those accent marks represent different vowel sounds, not stress shifts. Also, individual words can be emphasized just by saying them louder, but there's still room for a system of particles you attach to a word to put emphasis on them, which would carry over to writing.
Again, these are thoughts specifically on the kind of language the existing in-game constraints would imply, and "throw it away and make a new language" is still a viable alternative.