Your company consistently deals with both batch- and stream-based event data for various applications. To handle and process this data effectively, you intend to use Google Cloud Dataflow within a predictable time frame. Nevertheless, you have identified that, at times, the data might arrive later than expected or in a non-sequential manner. In light of these potential delays and disorder, how should you structure your Cloud Dataflow pipeline to appropriately manage and process data that arrives late or out of order?