You can accurately estimate this. This is not the first time projects have used new technologies or embarking on something that has never been done before.
Use all available techniques. Use a multidisciplinary team to provide the analysis and input. Do NOT use single-point estimates. Rather, have your team provide probabilistic values and assume a triangular distribution, i.e., optimistic, most likely, pessimistic.
Work top down, break your work down and go up, use Delphi technique, examine the extremes and eliminate, use historical values where you can, use parametrics where you can, and most of all iterate your work several times as you approach CONSENSUS.
When you break the work down, you will likely find much of the scope familiar and rather easy to estimate.
Once you have sequenced your work, simulate your schedule based on your probabilistic estimates and evaluate the results. Because this is new work, I would set my targets--the single point value that gets priced and baselined--on the heavy side of the curve; that is, I would target somewhere in the 80th to 90th percentile versus the more competitive 50th to 60th. By doing so, you are building in both schedule and cost contingencies in your targets.
You did not mention the type of contract under which you will perform this work. It is not appropriate for a firm fixed price vehicle. If you did that, you would add another contingency value and it would be huge for this type of risk, e.g., three to five times your price. It just does not make sense.
This is a T&M or cost plus type contract. You need to advise your client to build both management reserve--funds available to you to fund in scope but unplanned work--and contingency reserves for your client's use to fund anything else that bites him. The contingency reserves need to be rather healthy for work like this.
This is doable because you can rule out the extremes rather quickly. For example, from a top down perspective, will it take you ten years to do this project? Five? Three? Two? Do the same thing from the optimistic side. Will it take you a week? A month? Six month? Year? You can see how ten years and a week are laughable. When your team stops laughing, you have begun to establish either side of the probabilistic estimate.