Over time we accumulated a lot of features in elasticsearch core which have different status. Some of them are very important building blocks while other ones are only useful for esoteric use-cases or have known issues. Keeping the core small is important for a variety of reasons so maybe we should consider creating a new elasticsearch-extra plugin, with a best-effort only backward compatibility policy, which would contain such esoteric features.

Some ideas of what could move from core to such a plugin:

• the token_count field mapper (esoteric)
• the murmur3 field mapper (not really needed)
• the fuzzy_like_this query (super costly + esoteric)
• the top_children query (questionable approximation)
• the significant_terms aggregation (scalability issues on large data)
• the scripted_metric aggregation (useful for experimentation but hard to use and slow)