I think you should start by thinking about why age is important in your survey and identify what age intervals are meaningful for your analysis based on that answer. Do you have some hypotheses about why some age groups would be more or less satisfied with the yoga course? - Then you just need to measure to which of those groups each respondent belongs. Age is just a number, and if you think about it, it's pretty arbitrary: is there any magical difference between someone who is 36 and someone who is 37? No, but of course, age is often very meaningful because it's a proxy for many other important things in life. So, you mention the age of retirement; if you have a hypothesis about how retirees appreciate yoga classes more than working-age population because they are bored now that they don't go to work anymore, then yes, you absolutely want to make the age of 65 a cutoff point. Other such cutoff points could be for example the age at which people finish studies or get married. Other age groups that might like a yoga class more or less might be women who are likely to have given birth or women starting menopause (I'm just wild guessing here).
As to whether having uneven interval width of age groups is a problem from a statistical perspective - I say no, but that doesn't mean that there aren't consequences of how you choose to split your age variable. How you measure age will determine the type of variable you get, and that further determines your options for statistical analysis. If you have even intervals, you'll have yourself a proper "interval variable" where the difference between two values of the variable is numerically meaningful. If the intervals are uneven, you have a categorical variable, and if you only have two categories (e.g. retirees vs. non-retirees), a binary variable. If you're going to look at associations between variables or at statistical tests of independence, this determines what methods and tests you can do.
The one way you can have problems is if you want to compare groups, but don't have a lot of observations in some of them. If you define in advance some groups that you expect have very few people in them, you have to make sure that you'll get enough observations for whatever statistical analysis you want to do. You mention having few customers under the age of 18, so that could be a problematic group. And, of course, people have a tendency to - erh, die - as they get older, so old age groups can also be problematic if you have a small sample.
I hope this helps!
