I'm not intricately familiar with the NSF program, but I did see that the introduction to the standard acknowledges that (my emphasis): This manual provides criteria and expectations that the facility will be audited against and is generic for all types of food processing establishments. Some specific criteria may not be applicable. It is the responsibility of the manufacturer to justify that a specific criterion is not applicable.
My experience has been that a reasonable proportion of auditors are fairly knowledgeable and will intrinsically understand and accept that some criteria do not apply to the site being audited, and indeed this level of awareness and familiarity should really be a basic expectation of anyone conducting a paid-for third party audit, IMO. Alas that it isn't always the case though 
Exactly how you approach it may depend on the nature of your operations and indeed on the structure of your food safety management systems. If it's logical to include specific considerations relating to each clause within policy documents that relate to it then you could note it there.
Personally I'd be inclined to compile a specific document on the non-applicable clauses, detailing your reasoning behind each - that was you have something solid and considered that is readily to hand if/when auditors ask about it. The content may perhaps depend on the nature of the clause being omitted - this could be scientific data on the product/behaviour (for things like "we don't test x because published literature references a/b/c substantiate the position that it is not a risk for our product types), or simple operational considerations (e.g. "we don't have a procedure to calibrate our metal detector because we don't have a metal detector"). As long as your reasoning is sound and you can back it up I wouldn't expect any issue with this type of approach, and it's something I've used for parts of BRC. Nonetheless someone with more direct familiarity with NSF may be able to offer more specific guidance.