Mass customized production significantly complicates Computer-Aided Process Planning (CAPP), particularly when it comes to managing diverse manufacturing processes. In the current CAPP systems, process planning relies on the experience of the process planner. The latest achievements in using deep learning technologies for smart process planning systems have been limited by the data representation and the availability of labeled datasets. The contribution of the proposed paper is twofold: it provides a new method for the automated generation of the Bill of Process (BOP) based on graphical representation of the 3D-CAD-model and a new labeled dataset. The proposed method for automated generation of the BOP uses Graph Neural Network (GNN) and its inherent structure to extract the geometrical and informational attributes of the 3D-CAD-model. The components are represented as graphs, which encapsulates the relation between the machining features with their Product Manufacturing Information (PMI) such as surface finishes, shape and positional tolerances. To address the lack of accessible and adequate data, a new automatically generated dataset is proposed, which is built of manufacturable rotational parts with PMIs and machining features following the DIN-norms. The dataset is constructed of ISO-10303-242-STEP files and their corresponding BOPs. The presented GNN approach is trained and validated on the proposed dataset and the outcomes are examined. The resulting BOP details the operations for each machining feature, which are then grouped together into operation blocks. The outputted BOP is undisputed and manufacturable since each feature in the given part is assigned a specific operation.