My plan is to deprecate KDTrees.jl in favor of NearestNeighbors.jl. The KDTree in NN.jl is slightly faster and uses slightly less memory so there isn't really any point to KDTrees.jl anymore. Also, the tree in KDTrees.jl can only use Euclidian distance while the one in NN.jl can use the more general Minkowski distance.
I don't have any large plans to develop the package further (except to play around with threading support) but if anyone wants to use it as a sort of backend to another package I would (to the best of my ability) happily assist in solving any issues that arise.
