I would recommend taking an existing dataset core to Kedro like

pandas.CSVDataSet

and altering it for your purposes - since that's all tested to work https://kedro.readthedocs.io/en/stable/kedro.extras.datasets.pandas.CSVDataSet.html