Being able to split non-stationary time series into segments of similar stochastic behavior is an important task in signal processing with many applications from genomics to neurophysiology and finance. One approach to the problem consists of scanning the data by using moving sum statistics with multiple window sizes and to prune down the resulting candidate set to obtain final estimations for the location of changes. Most of the present work, however, deals with linear models (either mean-shifts or linear (auto-)regressive time series) while corresponding methods for non-linear time series are lacking. In this project, it is dealt with this problem class by evaluating moving sum statistics of estimating function securing detectability of all changes on the one hand and introducing new pruning techniques to obtain the final estimators on the other hand.