Based on the premise that a proform may only target a phrase, it has been well established that the proform <it> replaces NP (and not simply N). As shown by the following examples, adjuncts (as well as determiners) may not be left behind if a pronoun (or more accurately, a pro-NP) is used.
the red ball -> it
the red ball -> *the red it
the red ball -> *the it
the ball on the table -> it
the ball on the table -> *the it on the table
the ball on the table -> *it on the table
However, if the NP takes a complement (such as <of cake> and <of cheese> in the following examples), we see that the proform <one> replaces the noun and its complement but leaves behind the determiner and (optionally) any adjuncts (such as <big> and <small>). <one> seems to be replacing something other that what <it> replaces. Since <one> cannot replace just the noun when a complement is present, there must be some level of representation between N and NP. This is why we say that the proform <one> replaces N' (and not NP or N). The arguments below are daughters of N' and sisters of N, which means they cannot be left behind by proform replacement of N'. Adjuncts, however, can either be replaced by <one> or left behind as adjuncts on the proform.
the big piece and the small one
*the piece of cake and the one of cheese
the big piece of cake and the small one
*the big piece of cake and the one of cheese
*the big piece of cake and the small one of cheese
the piece on the table and the one in the fridge
*the piece of cake on the table and the one of cheese in the fridge
the big piece on the table and the small one in the fridge
*the big piece of cake on the table and the small one of cheese in the fridge
*the big piece of cake on the table and the one of cheese
Alternatively, we could analyze the data to mean that <one> is simply intransitive, like most nouns. We can't say <the one of cheese> because it's unclear how to interpret <of cheese> in this situation. However, <one> can certainly be <small>. As an adjunct, the meaning of <small> is less dependent on what it describes than the meaning of <of cheese> is on something like <piece>. In this view, the proform <one> is not really targeting something like <piece of cake>; it's targeting <piece> and rendering <of cake> ungrammatical. Notice that <of cake> can be dropped in the following example without its meaning slipping away. It is implied by <piece> the same way it is implied by <one> in the above examples.
the big piece of cake and the small piece
Although <one> can't take a complement such as <of cheese>, it can take <made of cheese>. Although the meaning is very similar to <of cheese> (just less vague), <made of cheese> must be analyzed as an adjunct to account for examples such as the following.
the wooden box and the one made of cheese
A polar distinction between complement and adjunct may be a little forced and arbitrary. I prefer to view them all as modifiers on sliding scale between complement and adjunct. A complement could be defined as a modifier that supplies necessary information about its head. The exact quality of this information is dependent on what head is being described. An adjunct, on the other hand, has an independent meaning that can apply consistently to virtually any head. There could be modifiers at all levels in between that apply to different percentages of heads and change their meaning to various degrees. X-bar theory suggests a very clear structural distinction between complement and adjunct.
One thing that still hasn't been cleared up with the current analysis is why the article <the> must be replaced by <it> but left behind by <one>. The two proforms certainly do replace different things. As I said, <one> can be analyzed as an intransitive noun (one that can take no complements). If <it> is analyzed as an intransitive determiner (instead of as a proform, ProDP), everything begins to fall into place. An example of a transitive determiner would of course be <the>. Noticing the parallels between verbs, prepositions, and determiners, it's not such a leap to call NP a complement of <the> in the following examples.
Because the entire NP falls under DP in this analysis, even NP adjuncts cannot be left behind when the transitive determiner <the> is replaced by an intransitive form such as <it>.
the ball -> it
the ball -> *the it
the red ball -> it
the red ball -> *the red it
the big red ball -> it
the big red ball -> *the big red it
the big ball of cheese -> it
the big ball of cheese -> *the big it
In this analysis, I see no reason why the N' level of NP (or its specifier position) is necessary. I'm sure there is a great deal of evidence of which I'm currently unaware, and I'd like to have a look at it and see if I can't analyze it without resorting to the complication of X'. If you'd like to argue with me in the comments, please feel free.