We're currently customizing the line breaking opportunities returned from ICU's line breaking iterator and rejecting those positions that would split an emoji sequence. 

This is work in progress in the Unicode consortium for defining the rules, and in ICU 57 on the implementation side. aheninger@google.com has been working on the relevant ICU changes, related ICU bug: http://bugs.icu-project.org/trac/ticket/12081

Once the required changes are in ICU, we should roll to that version and remove the custom code.