Ideally you wait until a number of stories and epics have passed through your Kanban board and then use that data to help estimate future cycle times.
If you absolutely need to provide estimates from day one then you can use the following approach:
- Break the stories down in to sub-tasks
- Estimate the hours for each sub-task
- Calculate the capacity of the team (number in the team multiplied by the available hours per person)
- Use this data to give some initial estimates of cycle time
This is similar to traditional development estimating and as such is vulnerable to over-optimism, pressure to size estimates based on business needs, etc. There is also an argument that this kind of estimating is a form of waste as it takes the team time to do but provides only limited value.
I would recommend that if you go with this approach that you drop it as soon as you have a reasonable amount of cycle time data available (say after a few weeks or a month).