I'm not quite sure if "load order independent" is the right term for them, but in looking at the keywords for http://www.nexusmods.com/skyrim/mods/49791/? and http://www.nexusmods.com/skyrim/mods/34093/? within TES5Edit, I noticed that the form ids were the same for both mods. This seems to be intended for the mods to function properly as seen in the http://forums.nexusmods.com/index.php?showtopic=1355223 and in an http://www.nexusmods.com/skyrim/articles/1420/? about Weapon and Armor Fixes Remade, both written by kryptopyr. The instructions point to using TES5Edit's Copy as Override Into feature to add keywords to your own mods to enable support for CCO and WAF without the need for them as plugin masters.

This leads to my question. How does one go about creating Keywords that are load order independent? Recently I figured out how to add race specific dialogue back for custom races that use Expired6978 and TMPhoenix's http://www.nexusmods.com/skyrim/mods/24168/?. I experimented by creating a patch file for Arthmoor's http://www.nexusmods.com/skyrim/mods/9557/? and I confirm that it does indeed work. The fix is simple, you just need to add conditionals in the dialogue topics for the proxy races that RaceCompatibility.esm provides. Of course, in order to access these keywords RaceCompatibility.esm needs to be a master for the patch. This is all fine and dandy as the only people who need support for custom races should be using the esm file anyway. However, it would really be great if these keywords could exist independently from the esm, so that authors could add support for custom races without the need for a patch. While the fix is simple, it is also extremely tedious to go through the dialogue topics in TES5Edit and look for race specific dialogue. This would not be a problem if the fix was implemented from the start.

If people are interested in how to re enable race specific dialogue for custom races. I'll add other post that describes the process in more detail.

EDIT: I pm'd krypotpyr about it and learned about injected records. Which is essentially what is asked about here. Apparently the method comes with a few caveats so it'll have be tested out to find out whether or not this is a good idea. I'm curious as to how many people would be interested in such a fix.