I suspect that this one would be a nightmare beyond reckoning to try to achieve.
Natural language is kind of an ugly business to begin with, so Unicode always had quite a task ahead of them; but constructing a defined algebra for combining pictograms is vastly different than doing one for even a fairly sprawling character based language or a use case with pretty intricate combinations(like mathematical typesetting or calligraphic scripts). The latter can be complex; and may leave you defining combinations that have no font support unless a natural language with a user base calls for it(eg. the ability to add metal umlauts to letters that don’t normally have them demands comparatively little from your font; the ability to have a sequence that never occurs in calligraphic script still flow nicely is more challenging); but you can cover a lot of ground without defining too many distinct combination operations; and a reasonable number of them are amenable to programmatic interpretation(like extending brackets to include enclosing NxM blocks of characters as required by matrix notation).
This doesn’t make them easy(Y’know what’s even cooler than trying to localize to support both left-to-right and right to left? The possibilities of bi-directional text, and LTR and RTL markers and overrides!); but at least being hard rather than nigh impossible means you can start work.
With pictograms, Not So Much.
In the case of the skin-color support you mention, say, “EMOJI MODIFIER FITZPATRICK TYPE” isn’t a special case of some color/tinting system, it was specifically thrown at the problem of representing human skin types so it is only considered a valid combination with 102 emoji, those deemed sufficiently human. There is no way to infer which are or aren’t on the list; there’s just a list(indeed, the list of miscellaneous symbols that are ‘emoji’ rather than just dingbats or miscellaneous symbols., is itself a category one can only know by consulting the list). For anything on the list, it’s meaningful; for anything not on the list, it isn’t(and, as far as I know, there isn’t a general color/tint system in Unicode because specifying text color isn’t supposed to be a character encoding’s job; so you can’t specify what non-human colors you would prefer for your human faces; nor can you distinguish between yellow, chocolate, and black labs) In order to add support for skin tone, Unicode now includes the concept of ‘humanity’ and enumerations of what is and isn’t part of it. Nasty epistemological scope-creep there.
That’s the trouble with pictograms. Because their appearance is supposed to be meaningful, and often closely related to real-world objects rather than abstracted letter forms, you can’t meaningfully define combination/relationship/modifier operations without also undertaking the Celestial Emporium of Benevolent Knowledge-class task of defining what categories things go in, what relations those categories can meaningfully exist in(and with which other categories); and whether the category implies the relationship(eg. you mentioned trains: do ‘things-that-are-train-cars’ connect together like trains when linked by a ZWJ; or is there a ‘connect in the manner of a train car’ joiner(and is it distinct from the ‘connect in the manner of a trailer’ joiner, or is that a modifier?) that causes any emoji that are not members of categories inimical to connecting in the manner of train cars to do so?).
For the most part, Emoji only escape the full brunt of this complexity by ignoring it: ‘these things are related’ all gets lumped into the zero width joiner(with limited variety handled by 16 variation selectors VS1-16, of which only a few seem to be used; none of them well defined because they are just being borrowed in this context); and what categories various glyphs are in isn’t represented anywhere in the encoding, you just have to hit the lists. What you get is both highly inconsistent(‘hot beverage’ is a thing, but it’s a single code point rather than “liquid + edible( + for humans) + serving (+ single) + hot (+ in the thermal sense)”; and even if you aren’t being that ridiculous I don’t think that there is a general set of temperature modifiers, despite numerous references to both heat and cold); heavily laden with assumptions ( “U+1F468, Man” + “U+2695 Staff of Aesculapius” is conventionally rendered as a doctor rather than a patient; despite their being numerous relationships between a human and the general concept of healing); and incomplete.
This still probably beats just carving out a new codepoint for all those combinations; but it adds a fun layer of epistemology to what was already a hopelessly broad mission statement in ways that having some control characters to allow you to superimpose glyphs or change text direction doesn’t.
.)
