The Machine Learning (ML) library for Flink is a new effort to bring scalable ML tools to the Flink community in order to make Flink a premier platform for ML. This effort covers a wide area of ML related activities including data analysis, model training and model serving. In this talk we will cover approaches to using Flink for model serving. We will present a framework and actual code for serving and updating models in real time. Although presented examples cover only two model types - Tensorflow and PMML, the framework is applicable for any type of externalizable models. We will also show how this implementation fits into overall ML learning implementation and demonstrate integration with SparkML and Tensorflow/Keras and specifically show how models can be exported from these systems. This initial implementation can be further enhanced to support additional model representations, for example PFA and introduction of model serving specific semantics into Flink framework.